List your company

Language Modeling on WikiText-2 (test) (Llama-3.1-8B-Instruct)

7.2PerplexityBF16 Baseline
1.56239.618577.675115.7315Jul 24, 2025

Evaluation results

MethodLinks
BF16 BaselineAvg. Bitwidth=16.0002025.077.2
HPTQAvg. Bitwidth=4.1252025.077.37
SSQR-5%Avg. Bitwidth=5.7252025.077.38
SSQR-4%Avg. Bitwidth=5.4052025.077.39
SSQR-3%Avg. Bitwidth=5.0852025.077.41
SSQR-2%Avg. Bitwidth=4.7652025.077.48
SSQR-1%Avg. Bitwidth=4.4452025.077.5
GPTQAvg. Bitwidth=4.1252025.077.56
HPTQAvg. Bitwidth=3.1252025.077.84
SSQR-5%Avg. Bitwidth=4.7252025.077.98
SSQR-4%Avg. Bitwidth=4.4052025.078.01
SSQR-3%Avg. Bitwidth=4.0852025.078.16
SSQR-2%Avg. Bitwidth=3.7652025.078.32
SSQR-1%Avg. Bitwidth=3.4452025.078.67
GPTQAvg. Bitwidth=3.1252025.079.44
HPTQAvg. Bitwidth=2.1252025.0711.89
SSQR-5%Avg. Bitwidth=3.7252025.0712.12
SSQR-4%Avg. Bitwidth=3.4052025.0713.22
SSQR-3%Avg. Bitwidth=3.0852025.0717.27
SSQR-2%Avg. Bitwidth=2.7652025.0725.18
SSQR-1%Avg. Bitwidth=2.4452025.0757.26
GPTQAvg. Bitwidth=2.1252025.07148.15